Reference Integration Case Study

Adding independent pre-execution governance to CLARA.

CLARA is a Sentinel SCA design partner and reference integration. The current phase is shadow governance: Sentinel evaluates and records proposed CLARA actions, but Sentinel does not currently block CLARA execution.

The first phase is shadow governance from CLARA's execution side: CLARA submits proposed actions to Sentinel, Sentinel returns a decision and evidence, and blocking is introduced only after both teams understand the observed decision patterns.

Discuss Governance Pilot View Sentinel Validation

Challenge

CLARA could progress from reasoning into tool execution. It had internal decision logic and resource-conflict controls, but actions reaching the executor did not pass through an independent external admissibility layer.

Sentinel Role

Sentinel was introduced as the external governance boundary. It does not replace CLARA's reasoning; it evaluates whether proposed actions satisfy policy, authority, risk, and evidence requirements.

Deployment Model

Shadow governance creates an evidence base first. Enforcement can then be enabled selectively for action families where the decision pattern is understood and the operational risk justifies blocking.

Governance Flow

CLARA proposes. Sentinel governs. Execution remains separate.

1Agent intent

CLARA proposes a tool action from its own reasoning loop.

2Sentinel check

Sentinel receives the action as a governance request.

3Policy decision

Sentinel returns ALLOW, REVIEW, or DENY.

4Evidence

When routed through /analyze, the decision is signed and recorded for audit and replay.

5Selective enforcement

Blocking is introduced only after shadow results are understood.

Evaluation Evidence

Early test and shadow evaluations demonstrated that Sentinel did not treat every autonomous action the same.

read_url health checksALLOW low risk
write_file test actionsREVIEW 0.62

Status: Design partner · Reference integration · Shadow governance. Sentinel evaluates and records proposed CLARA actions in shadow mode. Sentinel does not currently block CLARA execution. This is not presented as a customer deployment.

What It Demonstrates

  • Independent governance can sit outside the autonomous system being governed.
  • Low-risk and medium-risk actions can produce different outcomes.
  • Shadow mode lets teams build operational evidence before turning on blocking.
  • The pattern applies to agent workforces, DevOps automation, finance, robotics, IoT, and customer-data workflows.

Current Limitations

This case study does not claim Sentinel has taken over CLARA's execution path or that every CLARA action is fully enforced through Sentinel today. It documents the reference integration pattern, the initial test and shadow evaluation outcomes, and the agreed first phase of shadow governance.

CLARA remains responsible for reasoning and proposing actions. Sentinel independently determines whether those actions satisfy the conditions required to proceed toward execution.